What I'd actually install this week.
The stack I use every day to save time, move faster, and stay ahead. Not a tool review. Not a list of everything that exists. Just what's working — for me, for my team, for the agents I share the stage with. All of it running today inside an eight-figure real estate group — not a theory list.
The deeper explainers live on my YouTube.
Subscribe to my YouTube →Jump to a section
- 01Wispr Flow
- 02Perplexity
- 03The pre-meeting prompt
- 04Perplexity Computer and/or Manus
- 05Claude
- 06Claude CoWork
- 07Claude Code
- 08The Stack under the tools
- 09OpenClaw on a Mac Mini
- 10The AI engineer
- 11Fathom
- 12Granola
- 13Plaud
- 14SuperGrok
- 15Gemini
- 16NotebookLM
- 17Obsidian
- 18Voqo AI
- 19GoList
- 20Disclosures
Wispr Flow
Talk three times faster than you type.
Watch: Claude + Wispr running my weekMost agents type slow and write worse. Wispr fixes both — voice into any app, on any device, cleaned up in real time. Email, CRM, Slack, Notion, the lot.
The headline things you can do
- Brief your team without typing
- Capture buyer notes the moment you leave the door
- Voice-note a whole strategy on the drive home
The buy-back
Sixty to ninety minutes a day, week one. That's a listing presentation's worth of time, back in your diary.
Live web search with citations, on top of every major model. The closest thing to having a researcher on call.
The headline things you can do
- Pre-meeting briefs on vendors, buyers, agents, partners
- Live market research before you quote
- Reading and summarising long documents in minutes
Walk into every meeting prepared.
The single highest-leverage prompt in my stack. Drop into Perplexity. Paste in what you know. Get a structured brief in ninety seconds — communication style, decision style, values overlap, how to land the yes.
The ask-back rule: if you only give it a name, Perplexity will stop and ask for one anchor (address, social handle, mutual contact, or company) before it researches. No anchor, no answer.
See me run this live on a real meeting prepPerplexity Computer and/or Manus
Not just answers. Finished work.
Most AI you know is a chat window. These are different. You give them an outcome — go do this thing — they plan it, run sub-agents in parallel, connect to your tools, and deliver the finished result.
I run Perplexity Computer on Max (~AUD $300/mo) and/or Manus depending on the job. Same category, slightly different strengths. Manus is faster on creative, content and ads. Perplexity is sharper on research-heavy builds.
Real things I've actually run with these agents
See five real briefs I've run↓
The PA task I'd never get around to.
A custom website for a new listing.
A team retreat micro-site.
A whole Meta ads campaign.
A LinkedIn carousel about team learning.
The pattern. Anything you'd hire a freelancer or assistant for, and brief over Slack — these agents can do.
Watch: how I run these agents in a $15M companyWhere Perplexity is best for live research, Claude is best for thinking, writing, and reasoning through hard problems. Long documents, contracts, transcripts — drop the lot in, have a real conversation about it.
The headline things you can do
- Long-form content — posts, scripts, proposals
- Reviewing contracts and complex documents
- Working through strategy with a thinking partner
New to Claude? Set it up properly first. This is the exact setup I use, with the prompts. The video walks you through it — this page is the reference you bookmark. I keep it updated as the tools change.
Show the full Claude setup guide↓
1 · Get on it the right way
One Claude subscription gives you three things: Chat (thinking and writing), CoWork (it does the work), and Code (you build tools). Same login.
- Buy on the web, not the App Store. Subscribe at claude.com and download the desktop app. App Store in-app purchases can cost more and muddy your usage later.
- Email: use your Gmail or work email. Heading for a Team plan later? Start on a work email.
- Plan (AUD, at time of writing): Free to test → Pro ~$34/mo ($28 on annual) → Max from ~$170/mo (5×) if you hit limits → Team ~$43/seat. Add 10% GST unless you have an ABN. Limits are compute on a rolling 5-hour window, so big files and Opus burn faster. Pay monthly — it costs a touch more but keeps you free to switch as the tools change.
2 · Bring your ChatGPT brain across
In ChatGPT: Settings → Data controls → Export data (you get a conversations.json). In Claude: Settings → Capabilities → Memory → Start Import to pull your ChatGPT memory in. It's experimental, so check what carried over.
3 · Teach it your business
Claude is only as good as what it knows about you. Give it your brand voice, elevator pitches, ideal client profile (ICP), and brand colours — described in words, not just a logo file.
Where it goes (this trips people up): big-picture stuff that's true everywhere → account-level personalisation in Settings. Context for one slice of the business → that Project's instructions and knowledge. Abilities → Skills. Tools it plugs into → Connectors / MCP. Your actual knowledge base → markdown files you keep in Obsidian and feed into Projects.
No brand voice yet? Either get Claude to interview you (prompt below), or do Allan Dib's Lean Marketing One-Page Marketing Plan, save the output as a markdown file, and bring it in.
Act as a brand strategist. Interview me one question at a time to capture my brand voice, elevator pitch, ideal client profile (ICP), and core offers. Keep going until you're 95% confident you could write as me, then output it as one markdown brief I can save.
4 · Use Projects as a living RAG
A Project is a folder with a permanent memory. RAG (Retrieval-Augmented Generation) just means Claude reads your own files before it answers, instead of guessing from memory. Keep a markdown file per part of your business and update the file when things change. Anything you explain twice goes in the file. Claude is your second RAM; your files are the brain.
Avoid: dumping stacks of PDFs and asking it to read them every time — PDFs are heavy and torch your usage. Convert the important stuff to markdown.
Why markdown in Obsidian, not locked inside Claude: keep your brain portable and you can move between Claude, ChatGPT or whatever's best next. Claude's my favourite and the best place to start — it's not a cage. The reason I start people on Claude is the ecosystem: more people build on it, so you can pull skills, plug-ins and MCPs from just about anywhere.
5 · Tune how it thinks
- Make it disagree with you. Save a custom Style that tells it to challenge you, argue the other side, ask questions before big tasks, and cut the hype (prompt below).
- Models: Haiku (fast, cheap), Sonnet (daily driver), Opus (deep thinking). Sonnet for most, Opus when it matters, Haiku for speed.
- Skills: Settings → Feature Previews → turn on Code Execution first, then Skills. Built-ins worth having: Excel, PowerPoint, Word, PDF. Manage them in Customize → Skills.
Speak to me in plain Australian English. Be direct, no hype. Challenge my thinking — argue the other side, don't just agree. Ask clarifying questions before long tasks. Match my voice: short sentences, confident.
6 · Connect it to your tools
Customize → Connectors → enable Gmail, Drive and Calendar from the marketplace. Now Claude can read your real email and files.
7 · Custom MCP — when the connector doesn't exist
MCP (Model Context Protocol) is a universal adapter that lets Claude plug into any tool. If yours isn't in the marketplace: Customize → Connectors → "+" → add its name and remote MCP server URL. Find MCP servers in the tool's own docs or the public MCP directories.
On company Microsoft 365? Only the owner can add connectors at the org level. Don't build a workaround — ask your IT or managed service provider to add it properly and securely.
8 · What to actually do with it
Start where the business feels heavy and repetitive, not where the idea sounds exciting. Day-one wins: a daily brief off your real calendar and inbox; a client's full history pulled before a call; voice notes turned into clean docs in your voice.
Avoid: buying through the App Store, dumping PDFs, and treating it like a search box.
Claude with hands. You give it a task, walk away, come back to a finished result. Plans the work, runs sub-tasks, organises files, drafts, builds.
ChatGPT writes text. CoWork does the work. That's the difference.
Real things I've used it for
- Custom client portals for buyer advocacy clients to review shortlisted properties
- One-of-one proposals from a discovery call transcript — twelve minutes, in our brand
- Agency agreements drafted and auto-filled
- Internal training material captured, organised, hosted in one searchable place
- Inbox triage — surfaces what needs me, drafts replies to what doesn't
- File and folder management — the boring work that makes a business run
Claude Code writes and ships software from your terminal. You don't need to know how to code — you need to know what you want. This is how you stop paying SaaS rent and start owning your software.
What I'm building with it
- Custom dashboards for our Fishing Upstream pipeline — off-the-shelf CRMs don't fit how we run
- A personal brand content management system — ideas, carousels, clips, captions, branded automatically, published across every platform on schedule, performance scraped, fed back into the next round. Currently in testing. Building in public.
- Raw-to-Ready — the content pipeline. Raw footage in, finished short-form out, scheduled across every platform. Runs while I'm on appointments.
How I run it all in one system.
The layer under the tools.
Individual tools are cheap. What makes them compound is the layer that ties them together. This is what sits under mine.
What's under the tools
- Claude and OpenClaw running in unison. Claude for the thinking. OpenClaw for the reach — data sources most agents can't touch, hooked into a private toolkit. A small set of other models rounds it out where they earn their place.
- Hosted agents in AWS, private environment. The heavier workloads don't live on a laptop. They run in our own AWS environment — private data, private tools, private context. Nothing shared with model providers we don't want shared.
- In-house live dashboards. One pane of glass across every online activity that matters. Content performance, pipeline health, market intel, agent capacity — all live, all ours. No SaaS rent, no vendor lock-in.
- A business operating system the whole team works to. Updated by voice on a daily rhythm. When someone learns something, it lands in the system that afternoon. The team doesn't chase context. The system holds it.
- The Raw-to-Ready content pipeline. Raw footage in, finished short-form out. Transcribed, cut, captioned, branded, scheduled across every platform. What used to be a full editor's job now runs while I'm on appointments.
None of this replaces the tools above. It's what makes them worth having.
OpenClaw on a Mac Mini
My always-on research engine.
A Mac Mini in the corner of the office. Running an open-source build of Claude with browser-use and a custom mix of search and scraping tools. Toolset is essentially infinite — whatever data needs reaching, the rig can be set up to reach it.
What the rig actually does↓
What it does for me
- Scrapes development site opportunities across every council DA register in our service area
- Builds buyer advocacy shortlists from portals, off-market and pre-market intel
- Scans for future agents and team members — content, public reviews, recent moves
- Ambient competitive intelligence — what other agencies are doing, instead of frantic catch-up
It can also build software — same as Claude Code, but more in the dark. Claude Code shows me every file and decision. OpenClaw just builds it while I'm doing something else. Both have a place.
Same rig now runs in two places. A Mac Mini in the corner of the office for the always-on jobs, and a private AWS environment for the heavier workloads. Private data, private tools, private context.
This is the most advanced thing in my stack and the one most agents won't bother with. The ones who do bother will compound an unfair advantage.
The AI engineer.
The one hire that makes everything above run — without you becoming the technician.
Everything on this page compounds faster when someone qualified builds it into your business properly — inside your systems, on your data, with nothing sensitive going where it shouldn't. That's not another SaaS subscription. That's an AI engineer.
I partner with Andrew Herbert of Remap AI, backed by Private AI, for exactly this. Andrew builds and hosts the deeper agent infrastructure behind a lot of what's on this page — the private AWS environment, the hosted agents, the always-on rig. I'm the operator; Andrew is the engineer. What we're building together lives at In-House AI.
The 12-minute version
What an AI engineer actually does — and the four ways to hire one↓
What they actually do
Not prompt engineering. Not reselling ChatGPT. A qualified AI engineer designs and builds AI systems that integrate with your existing tools, protect your data, and keep running after the handover — custom agents, private hosting, internal automations your CRM has never heard of.
The four-question filter
Before any AI project spends a dollar, it has to pass four questions — I walk through them in the episode. If a project can't answer all four, it doesn't get built. That filter has saved me more than any tool on this page.
The four hiring models
DIY, BPO, done-with-you, and full agency. In the episode I compare real Australian and US costs across all four — including the $75K outsourced project that blew out on me, later rebuilt for a fraction of the cost by the right engineers.
Who this is for
- Businesses handling sensitive client data that can't go near a public LLM
- Teams who've hit the ceiling on off-the-shelf AI tools and need something custom
- Anyone quoted six months and $200K by a consultancy who suspects there's a better path
Joins Zoom, Teams or Meet automatically. Transcribes, summarises, flags the moments. The bit that makes it different — runs the conversation against proven sales frameworks and tells you exactly where the call hinged.
How I grade calls with BANT and SPICED↓
BANT — buyer discovery and pipeline calls
Budget, Authority, Need, Timeline. Use it on every buyer enquiry, every discovery call, every "are they actually serious" conversation. You stop guessing where they sit in the pipeline. You know.
SPICED — the closing conversations
Situation, Pain, Impact, Critical Event, Decision. Listing presentations, buyers advocacy, property management onboarding, off-the-plan. SPICED tells you whether you're hitting the buttons that convert.
No bot in the attendee list. No banner. No moment where you have to explain why a robot just joined the vendor meeting. You type rough notes, Granola weaves them with the transcript afterwards.
A small recorder around your neck or pinned to your lapel. Catches the conversations, ideas and musings that used to disappear the moment they happened.
What I use it for
- Open home walk-throughs and in-car debriefs after appointments
- Musings while driving — half-formed ideas, decisions, content angles
- Notes to my team and notes to myself, dictated as they come
Plugged directly into X. Best in the stack for live data, current events and the conversation as it's happening. Also the best image and image-to-video pipeline I've found for ad creative.
Gemini
I don't love the interface. But Nano Banana is too good to leave out.
Gemini's image model is best-in-class. The chat experience isn't where I want to think. So I reach Gemini through other tools via API or MCP — Perplexity calls it for some queries, Claude can be wired to it through MCP. The image model gets used; the chat window doesn't.
Drop in your sources — documents, transcripts, web pages — and ask anything. Every answer is cited back to the source paragraph. Also has Google's AI-podcast feature, which produces a remarkably good two-host discussion of whatever you give it.
I mostly wire it into a stack via MCP rather than using it directly. For agents not building stacks — still a great standalone for working through long documents.
Most people's best ideas disappear into apps they never open again. Obsidian is the opposite — a local, private, permanently yours knowledge base built entirely in plain text. Notes link to each other. The more you build, the smarter it gets.
I use it to capture and connect the frameworks, systems and half-formed thinking that run my business. It's where the raw ideas from a walk, a drive or a meeting get turned into something I can actually build with. Presence OS, Fishing Upstream, The Ownership Blueprint — the thinking behind all of it started here.
What makes Obsidian different↓
The headline things you can do
- Build a second brain that compounds over time — every idea connected to every other relevant idea
- Wire your AI tools into it — Claude, Perplexity and NotebookLM can all read from and write into Obsidian via plugins
- Work completely offline, completely private — nothing in the cloud unless you choose it
The bit that makes it different
Most knowledge tools are search boxes. Obsidian shows you the graph — how your thinking connects. Once you've been building in it for a while, patterns emerge that you couldn't have seen in a list.
It's free. It's fast. It doesn't require a subscription to keep what you've built.
Replaces voicemail with a real conversation. When you can't answer, Voqo picks up — qualifies the lead, books inspections, hands off to a human when it matters, pushes the brief to your CRM.
Running across our leasing line right now. Five LGAs in its head. Knows our pre-market stock. When you enquire, mention Mark Kentwell to let them know how you found them.
Consolidates leads from every source — portals, your site, paid ads, referrals — and routes them to the right agent within seconds. Companion product GoConnect captures the calls and texts that follow.
The difference between a five-minute response and a fifty-minute response is the entire conversion rate.
I'd rather be honest than impressive.
GoList — I sit on their Product Advisory Panel. I'm also an investor.
Voqo AI — I sit on their Customer Advisory Panel.
Wispr Flow, Fathom, Granola, Plaud — referral links. To be honest, you'll probably get more benefit from the referral than I will.
I work in partnership with Andrew Herbert of Remap AI, backed by Private AI. Andrew builds and hosts the deeper agent infrastructure that sits behind a lot of what's on this page. If you want a look at what we're building together, that lives at In-House AI. If I connect a business to an engineer through this arrangement, I may receive a referral fee — you’ll always know when that’s the case.
I'm a foundation investor in PieLab Capital, an Australian investment group running two funds.
The first is the Real Estate Industry Venture Capital (REIVC) Fund — closed for further investment, built around seven Australian proptech companies: ActivePipe, AiRE (RiTA), Cloudstaff, HappyCo, Detector Inspector, SATS (Smoke Alarm Testing Services) and :Different. I had equity in all seven. ActivePipe was acquired by MoxiWorks in 2022. AiRE was acquired by CoreLogic and now sits inside Cotality. I still hold equity in SATS and HappyCo today.
The second is the Council Capital Fund — PieLab's flagship buy-and-hold serial acquirer in property services and adjacent sectors. No direct proptech in this fund.
Beyond PieLab, I'm an active investor in other proptech and SaaS tools.
When I endorse something here, it's because I genuinely believe in the product. I don't invest cold. Most of the time I've been a customer for a long time before any equity conversation, and I've used what's on this page in my own business before recommending it to anyone else.
If a tool appears here without a called-out relationship, assume there isn't one. If one develops later, I'll update the page.
This is just a starting point.
At any time I'm testing thirty-plus AI tools. This page is the today version — the ones I'd genuinely tell another agent to install this week.
The real learning happens elsewhere. I post daily on Instagram. I go deeper on YouTube — full breakdowns, build-in-public episodes, the longer conversations. If today gave you something useful, follow along. The compounding starts when you stop watching and start using.
Recent episodes worth your time
One more thing. Everything on this page runs live inside Presence — my team works with this stack every day, and it's a big part of how we're scaling across the country without burning people out. If you're an agent or operator who'd rather work at this level than just read about it, the door's open. Start on YouTube, or say hello on LinkedIn.
